{
  "family": "vietneu_tts",
  "display_name": "VieNeu-TTS v3 Turbo",
  "description": "On-device Vietnamese TTS model with instant voice cloning from 3-5 seconds of reference audio, English-Vietnamese code-switching, streaming playback, batched generation, and conversation mode.",
  "category": "tts",
  "status": "community",
  "tasks": [
    "tts",
    "clone"
  ],
  "modes": [
    "offline"
  ],
  "languages": [
    "vi",
    "en"
  ],
  "capabilities": {
    "clone": [
      "speaker_reference"
    ]
  },
  "options": {
    "request": [
      {
        "name": "reference_text",
        "type": "string",
        "description": "Transcript matching the reference speaker WAV for ICL voice cloning.",
        "required": false
      },
      {
        "name": "speaker_embedding_path",
        "type": "path",
        "description": "Path to a speaker embedding .emb.txt file containing 192 comma-separated float values.",
        "required": false
      },
      {
        "name": "speaker_embedding",
        "type": "float_list",
        "description": "Comma-separated list of 192 speaker embedding float values.",
        "required": false
      },
      {
        "name": "speaker_embedding_only",
        "type": "bool",
        "description": "Use only the speaker embedding for voice cloning instead of ICL reference codec prompting; default false.",
        "required": false,
        "default": false
      },
      {
        "name": "max_tokens",
        "type": "int",
        "description": "Maximum generated codec frames; default comes from generation_config.json, 2048 when absent.",
        "required": false,
        "min": 1,
        "default": 2048
      },
      {
        "name": "do_sample",
        "type": "bool",
        "description": "Sample the main codec stream instead of greedy decoding; default true.",
        "required": false,
        "default": true
      },
      {
        "name": "temperature",
        "type": "float",
        "description": "Main codec sampling temperature; default 0.9.",
        "required": false,
        "default": 0.9
      },
      {
        "name": "top_k",
        "type": "int",
        "description": "Main codec top-k sampling limit; default 50, 0 disables top-k filtering.",
        "required": false,
        "min": 0,
        "default": 50
      },
      {
        "name": "top_p",
        "type": "float",
        "description": "Main codec nucleus sampling limit; default 1.0.",
        "required": false,
        "min": 0.0,
        "max": 1.0,
        "default": 1.0
      },
      {
        "name": "repetition_penalty",
        "type": "float",
        "description": "Positive repetition penalty for the main codec stream; default 1.05.",
        "required": false,
        "default": 1.05
      },
      {
        "name": "subtalker_temperature",
        "type": "float",
        "description": "Acoustic decoder sampling temperature; default 0.9.",
        "required": false,
        "default": 0.9
      },
      {
        "name": "subtalker_do_sample",
        "type": "bool",
        "description": "Enable sampling in the acoustic decoder; default true.",
        "required": false,
        "default": true
      },
      {
        "name": "subtalker_top_k",
        "type": "int",
        "description": "Acoustic decoder top-k sampling limit; default 50, 0 disables top-k filtering.",
        "required": false,
        "min": 0,
        "default": 50
      },
      {
        "name": "subtalker_top_p",
        "type": "float",
        "description": "Acoustic decoder nucleus sampling limit; default 1.0.",
        "required": false,
        "min": 0.0,
        "max": 1.0,
        "default": 1.0
      },
      {
        "name": "seed",
        "type": "int",
        "description": "Codec sampling seed; omitted requests choose a random seed.",
        "required": false,
        "min": 0
      },
      {
        "name": "text_chunk_size",
        "type": "int",
        "description": "Maximum UTF-8 codepoints per generated text chunk; default 200.",
        "required": false,
        "min": 1,
        "default": 200
      },
      {
        "name": "text_chunk_mode",
        "type": "enum",
        "description": "Framework text chunking mode; default word_budget.",
        "values": [
          "word_budget",
          "endline",
          "tag_aware"
        ],
        "required": false,
        "default": "word_budget"
      }
    ],
    "session": [
      {
        "name": "mem_saver",
        "type": "bool",
        "description": "Release the talker cached-step graph after each request; default false.",
        "required": false,
        "default": false
      },
      {
        "name": "voice_prompt_cache_slots",
        "type": "int",
        "description": "Voice prompt cache slots; default 1.",
        "required": false,
        "min": 0,
        "default": 1
      },
      {
        "name": "weight_type",
        "type": "enum",
        "description": "Shared talker and voice-prompt matmul weight storage type; default native.",
        "preset": "weight_type_full",
        "required": false,
        "default": "native"
      },
      {
        "name": "talker_weight_type",
        "type": "enum",
        "description": "Talker weight storage type; defaults to weight_type when set, otherwise native.",
        "preset": "weight_type_full",
        "required": false
      },
      {
        "name": "speech_encoder_weight_type",
        "type": "enum",
        "description": "Reference speech encoder matmul weight storage type; default native.",
        "preset": "weight_type_full",
        "required": false,
        "default": "native"
      },
      {
        "name": "speech_decoder_weight_type",
        "type": "enum",
        "description": "Speech decoder matmul weight storage type; default native.",
        "preset": "weight_type_full",
        "required": false,
        "default": "native"
      },
      {
        "name": "conv_weight_type",
        "type": "enum",
        "description": "Speech tokenizer and speaker-encoder convolution weight storage type; default f32.",
        "preset": "weight_type_conv",
        "required": false,
        "default": "f32"
      },
      {
        "name": "talker_graph_arena_mb",
        "type": "int",
        "description": "Talker graph arena size in MiB; default 256.",
        "required": false,
        "min": 1,
        "default": 256
      },
      {
        "name": "speech_encoder_graph_arena_mb",
        "type": "int",
        "description": "Reference speech encoder graph arena size in MiB; default 32.",
        "required": false,
        "min": 1,
        "default": 32
      },
      {
        "name": "speech_decoder_graph_arena_mb",
        "type": "int",
        "description": "Speech decoder graph arena size in MiB; default 32.",
        "required": false,
        "min": 1,
        "default": 32
      },
      {
        "name": "speaker_encoder_graph_arena_mb",
        "type": "int",
        "description": "Speaker encoder graph arena size in MiB; default 32.",
        "required": false,
        "min": 1,
        "default": 32
      },
      {
        "name": "talker_constant_context_mb",
        "type": "int",
        "description": "Talker constant tensor context size in MiB; default 4096.",
        "required": false,
        "min": 1,
        "default": 4096
      },
      {
        "name": "code_predictor_constant_context_mb",
        "type": "int",
        "description": "Code predictor constant tensor context size in MiB; default 1536.",
        "required": false,
        "min": 1,
        "default": 1536
      },
      {
        "name": "speech_decoder_constant_context_mb",
        "type": "int",
        "description": "Speech decoder constant tensor context size in MiB; default 1536.",
        "required": false,
        "min": 1,
        "default": 1536
      }
    ],
    "load": []
  },
  "runtime": {
    "tags": [
      "gguf"
    ]
  },
  "sources": [
    {
      "format": "gguf",
      "roots": {
        "model": ".",
        "weights": "$gguf"
      },
      "files": {
        "config": "model:config.json",
        "tokenizer_config": "model:tokenizer_config.json",
        "speech_tokenizer_config": "model:speech_tokenizer/config.json"
      },
      "optional_files": {
        "generation_config": "model:generation_config.json",
        "vocab": "model:vocab.json",
        "merges": "model:merges.txt",
        "tokenizer_json": "model:tokenizer.json",
        "special_tokens_map": "model:special_tokens_map.json"
      },
      "tensors": {
        "model_weights": {
          "source": "weights:",
          "prefix": "model_weights"
        },
        "speech_tokenizer_weights": {
          "source": "weights:",
          "prefix": "speech_tokenizer_weights"
        }
      }
    },
    {
      "format": "safetensors",
      "roots": {
        "model": "."
      },
      "files": {
        "config": "model:config.json",
        "tokenizer_config": "model:tokenizer_config.json",
        "speech_tokenizer_config": "model:speech_tokenizer/config.json"
      },
      "optional_files": {
        "generation_config": "model:generation_config.json",
        "vocab": "model:vocab.json",
        "merges": "model:merges.txt",
        "tokenizer_json": "model:tokenizer.json",
        "special_tokens_map": "model:special_tokens_map.json"
      },
      "tensors": {
        "model_weights": "model:model.safetensors",
        "speech_tokenizer_weights": "model:speech_tokenizer/model.safetensors"
      }
    }
  ],
  "packages": [
    {
      "id": "vietneu_tts_v3_turbo_q8_0",
      "display_name": "VieNeu-TTS v3 Turbo GGUF",
      "default": true,
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "VieNeu-TTS-v3-Turbo-GGUF",
      "files": [
        "model.gguf"
      ],
      "strip_prefix": ".",
      "download": {
        "kind": "huggingface_snapshot",
        "repo": "phuocnguyen90/VieNeu-TTS-v3-Turbo-GGUF"
      }
    }
  ]
}